Spine, posture and sole integrated detection method

Abstract

The invention discloses a spine, posture and sole integrated detection method, and belongs to the field of image processing. Image capturing is carried out on the spine, the posture and the pelma, algorithm processing is carried out through the control host, finally, data information is synchronized to the cloud platform, doctor management is facilitated, and a patient database is formed. The technology can effectively improve prevention and tracking treatment effects on spine, posture and sole health of teenagers in practical application scenes. The system is helpful for improving the efficiency of teenager health screening detection and the controllability of the physical examination process, and plays an important role in the medical health industry.

Description

Technical Field

[0001] The invention belongs to the field of image processing, in particular to a method for integrated detection of spine, posture and sole. Background Art

[0002] Spine, posture and sole health detection equipment helps improve the efficiency of adolescent health screening and plays an important role in the medical and health industry. However, traditional spine health screening instruments, posture health screening instruments and sole health screening instruments have a single detection method and few detection indicators. In the actual adolescent health screening process, traditional instruments cannot meet the needs of measuring multiple indicators at the same time to draw conclusions.

[0003] At present, the main detection methods at home and abroad include X-ray imaging, whole body three-dimensional imaging system (EOS), ultrasound detection, APP (photography + angle measurement), cloud pattern detection, etc. In the application of large-scale screening of adolescent spine health, X-rays and whole body three-dimensional imaging systems (EOS) have a certain radiation dose, which is harmful to the human body. During the screening process, ultrasound detection is slow and expensive, and is not suitable for large-scale censuses. Therefore, the above-mentioned detection methods generally have problems such as safety, reliability, ease of use, ease of operation, economic applicability, and difficulty in popularization and promotion.

[0004] Aiming at the needs of spine, posture and sole health detection for adolescents, a spine, posture and sole integrated detection method was proposed. Summary of the invention

[0005] In response to the problems of traditional youth health monitoring technology that cannot achieve intelligent filing and lacks comprehensiveness, this patent invents an integrated health monitoring system for the spine, posture and soles of the feet. The system captures images of the spine, posture and soles of the feet separately, and then uses the control host to perform algorithm processing. In actual application scenarios, the system can improve the prevention and follow-up treatment effects of the spine, posture and sole health of adolescents.

[0006] The technical solution adopted by the present invention is as follows: a method for integrated detection of spine, posture and sole, the method comprising:

[0007] Step 1: Use a depth camera to collect visible light images and 3D point cloud image data of the subject in multiple positions and angles;

[0008] Step 2: When the back of the person is facing the camera, start the foot scan at the same time;

[0009] Step 2.1: The subject stands still, places both feet in the scanner, and adjusts the position and angle;

[0010] Step 2.2: Obtain three-dimensional data of the foot and transfer the data to a computer;

[0011] Step 3: Acquire coronal and sagittal dynamic back images;

[0012] Step 3.1: Collect visible light and 3D images of the coronal plane and left and right lateral flexion;

[0013] Step 3.2: Acquire visible light and 3D images of sagittal flexion;

[0014] Step 4: Calculate relevant index parameters according to the collected images and output the calculated relevant parameters;

[0015] Step 4.1: YOLOv8 is used as the back coronal plane key point detection model. The collected images are annotated with annotation software, and divided into training set, validation set and test set. The annotated training set is trained to obtain the trained back coronal plane key point detection model. The key points detected by the back coronal plane key point detection model include: left and right tragus, left and right posterior axillary line vertices, left and right scapulae, left and right lumbar concavities, left and right iliac crests, left and right popliteal fossa, left and right ankles, left and right heels, the midpoint of the back neck C7, and the midpoint of the waist L5.

[0016] Step 4.1.1: Based on the key points of the head detected by the key point detection model, i.e., the left and right tragus points, the head tilt angle is measured according to the angle between the line connecting the two points and the horizontal line, and the tilt angle is output;

[0017] Step 4.1.2: According to the key points C7 and L5 detected by the key point detection model, the spine is segmented and evenly divided into the midpoint of the back neck C7, the intersection of the line connecting the inner superior angles of the two shoulder blades and the longitudinal axis of the spine, and the waist. According to the length of the segment, with C7 as the origin, the coordinates of the key points are searched upward, mapped to the 3D point cloud, and the lumbar curve is fitted according to the point cloud to calculate the cervical lordosis angle;

[0018] Step 4.1.3: Based on the key points of the left and right shoulder blade heights detected by the key point detection model, use the Y coordinates of the key points to perform height comparison, calculate the angle between the connecting line and the horizontal line, that is, the degree of inclination, and output the degree of inclination;

[0019] Step 4.1.4: According to the key points of the left and right scapula detected by the key point detection model, map them to the 3D point cloud according to their X and Y coordinates, obtain their 3D coordinates (X, Y, Z), calculate the scapula concavity according to the difference in the Z coordinates, and output the scapula concavity;

[0020] Step 4.1.5: Use the Canny operator to perform edge detection on the color image. According to the key points of the posterior axillary lines on both sides detected by the key point detection model, extend upward to make a Y-direction reference line. The intersection with the image contour is the acromion. According to the angle between the connecting line of the acromion key points and the horizontal line, calculate the inclination of the shoulder and output it;

[0021] Step 4.1.6: Calculate the symmetry of the distance from scapula to spinous process: According to the key points of the scapula on both sides detected by the key point detection model, calculate the distance from the key points of the scapula to the line connecting C7 and L5, obtain the symmetry of the distance from scapula to spinous process and output it;

[0022] Step 4.1.7: Based on the waist concave key points detected by the key point detection model, with the waist concave points on both sides as the center, and based on the contour image calculated in 4.1.5, search upward / downward to obtain the waist concave point set, perform Hough circle fitting on the waist concave on both sides respectively, calculate the curvature of the fitting circle, compare and obtain the waist concave symmetry index and output it;

[0023] Step 4.1.8: Based on the waist concave key points detected by the key point detection model, combined with the contour of 4.1.5, extend to the left waist concave point and the right waist concave point respectively, and the intersection with the contour is the reference point for calculating the distance from the waist concave to the arm. Calculate the distance from the reference points on both sides to the waist concave, compare and obtain the symmetry index of the distance from the waist concave to the arm and output it;

[0024] Step 4.1.9: According to the key points of the left and right iliac crests detected by the key point detection model, the angle between the connecting line and the horizontal line is calculated to obtain the inclination of the iliac crest and output it;

[0025] Step 4.1.10: Based on C7 and L5, train the Yolov8 key point model to obtain the 5 key points of the epidermal spinous process. According to the calculated 5 key points, fit the epidermal spinous process connection line, calculate the scoliosis angle and output it;

[0026] Step 4.1.11: According to the coordinates of the color map of the key points T1-T12 calculated in 4.1.2, map them to the 3D point cloud, obtain their corresponding X, Y, and Z coordinates, fit a circular arc in the plane formed by the Y and Z coordinates, calculate the kyphosis angle, and when the kyphosis angle is too large and exceeds 45 degrees, there is "excessive kyphosis", and output the detection result;

[0027] Step 4.1.12: Map the coordinates of the color image of the key points L1-L5 calculated in 4.1.2 to the 3D point cloud to obtain their corresponding X, Y, and Z coordinates, fit a circular arc in the plane formed by the Y and Z coordinates, and calculate the lordosis angle. If the lordosis angle is too large and exceeds 45 degrees, there is "excessive lordosis", and output the detection result;

[0028] Step 4.2: Use the YOLOv8 model as the frontal coronal plane key point detection model, annotate the collected images with annotation software, and divide them into training set, validation set and test set, train the annotated training set, and obtain the trained frontal coronal plane key point detection model; the key points detected by the frontal coronal plane key point detection model include: left and right greater trochanters of the femur, left and right knee joints, and left and right ankles;

[0029] Step 4.2.1: Leg length difference: Compare the Y coordinates of the key points of the greater trochanter of both sides to determine whether there is leg length difference and output the detection result;

[0030] Step 4.2.2: Knee valgus: Calculate the angle based on the line connecting the greater trochanter of the femur and the knee joint and the line connecting the knee joint and the ankle, and determine whether the angle is around 180°. If it is significantly less than 180°, it is knee valgus. On the contrary, if it is significantly greater than 180°, it is knee valgus, and the test result is output;

[0031] Step 4.3 Sagittal plane parameters: key point detection algorithm YOLOv8, collect images, annotate them with annotation software, and divide them into training set, validation set and test set. Train the annotated training set to obtain the trained key point detection model. The key points include: tragus, cervical 7 (C7), apex of posterior axillary line, popliteal fossa, ankle, heel and toe;

[0032] Step 4.3.1: Head forward: Based on the tragus key points and C7 detected by the key point detection model, calculate the angle between the line connecting the two key points and the horizontal line. If the angle is significantly smaller than that of a normal person, it is considered as head forward tilt and the detection result is output;

[0033] Step 4.3.2: Round shoulder: According to the round shoulder key points detected by the key point detection model, the Canny operator is used to detect the edge of the image. The round shoulder key points extend to the right along the X direction, and the intersection with the edge is used as the reference point. The distance from the reference point to the round shoulder point is calculated. If the distance is too large, it is a round shoulder, and the detection result is output;

[0034] Step 4.3.4: Knee hyperextension: Calculate the angle between the line connecting the key point of the popliteal fossa and the heel and the vertical line. If the angle is greater than 5°, it is judged as knee hyperextension and the detection result is output;

[0035] Step 4.3.5: Pointed feet: Based on the key points of the toes and heels detected by the key point detection model, measure the angle between their connecting line and the horizontal line. If the angle is too large, it is judged as pointed feet and the detection result is output.

[0036] Furthermore, the specific method of step 1 is:

[0037] Step 1.1: The subject stands in the center of the turntable, with his back facing the depth camera and fixed in an upright position;

[0038] Step 1.2: The depth camera collects visible light and 3D images;

[0039] Step 1.3: Rotate the turntable 90°;

[0040] Step 1.4: Repeat 1.2-1.3 until 360° three-dimensional image acquisition is completed.

[0041] In order to solve the problem that traditional youth detection technology cannot realize intelligent archiving and lacks comprehensiveness, the present invention invents an integrated detection system for spine, posture and sole. The system captures images of the spine, posture and sole respectively, and then uses the control host to perform algorithm processing, and finally synchronizes the data information to the cloud platform for doctors to manage and form a patient database. This technology can effectively improve the prevention and follow-up treatment effects of adolescent spine, posture and sole health in actual application scenarios. The system helps to improve the efficiency of adolescent health screening and the controllability of the physical examination process, and plays an important role in the medical and health industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the image acquisition flow chart.

[0043] Figure 2 Schematic diagram of the key points on the back.

[0044] Figure 3 Schematic diagram of the key points of the scoliosis angle measured on the back of the spinous process line.

[0045] Figure 4 Schematic diagram of the front key points.

[0046] Figure 5 Schematic diagram of key points on the side.

[0047] Figure 6 Schematic diagram of back detection indicators.

[0048] Figure 7 Schematic diagram of positive detection indicators.

[0049] Figure 8 Schematic diagram of side detection indicators.

[0050] Fig. 9 Schematic diagram of sagittal flexion detection indicators.

[0051] Fig.10 Schematic diagram of coronal plane lateral flexion symmetry detection indicators. DETAILED DESCRIPTION

[0052] The technical solution adopted by the present invention is as follows: a method for integrated detection of spine, posture and sole, the method comprising:

[0053] Step 1: Use a depth camera to collect visible light images and 3D point cloud image data of the subject in multiple positions and angles;

[0054] Step 2: When the back of the person is facing the camera, start the foot scan at the same time;

[0055] Step 2.1: The subject stands still, places both feet in the scanner, and adjusts the position and angle;

[0056] Step 2.2: Obtain three-dimensional data of the foot and transfer the data to a computer;

[0057] Step 3: Acquire coronal and sagittal dynamic back images;

[0058] Step 3.1: Collect visible light and 3D images of the coronal plane and left and right lateral flexion;

[0059] Step 3.2: Acquire visible light and 3D images of sagittal flexion;

[0060] Step 4: Calculate relevant index parameters based on the collected images;

[0061] Step 4.1: Use YOLOv8 as the back coronal plane key point detection model, annotate the collected images with annotation software, and divide them into training set, validation set and test set, train the annotated training set, and obtain the trained back coronal plane key point detection model; the key points detected by the back coronal plane key point detection model include: left and right tragus, left and right posterior axillary line apex, left and right scapula, left and right lumbar concavities, left and right iliac crests, left and right popliteal fossa, left and right ankles, left and right heels, the midpoint of the back neck C7, and the midpoint of the waist L5; Figure 2 shown.

[0062] Step 4.1.1: Based on the key points of the head detected by the key point detection model, i.e., the left and right tragus points, the head tilt angle is measured according to the angle between the line connecting the two points and the horizontal line;

[0063] Step 4.1.2: According to the key points C7 and L5 detected by the key point detection model, the spine is segmented and evenly divided into the midpoint of the back neck C7, the intersection of the line connecting the inner superior angles of the two shoulder blades and the longitudinal axis of the spine, and the waist. According to the length of the segment, with C7 as the origin, the coordinates of the key points are searched upward, mapped to the 3D point cloud, and the lumbar curve is fitted according to the point cloud to calculate the cervical lordosis angle;

[0064] Step 4.1.3: Based on the key points of the left and right shoulder blade heights detected by the key point detection model, use the Y coordinates of the key points to compare the heights and calculate the angle between their connecting line and the horizontal line, i.e. the degree of inclination;

[0065] Step 4.1.4: According to the key points of the left and right scapula detected by the key point detection model, map them to the 3D point cloud according to their X and Y coordinates, obtain their 3D coordinates (X, Y, Z), and calculate the concavity of the scapula according to the difference in the Z coordinates;

[0066] Step 4.1.5: Use the Canny operator to perform edge detection on the color image. According to the key points of the posterior axillary lines on both sides detected by the key point detection model, extend upward to make a Y-direction reference line. The intersection with the image contour is the acromion. According to the angle between the line connecting the acromion key points and the horizontal line, calculate the degree of shoulder inclination;

[0067] Step 4.1.6: Calculate the symmetry of the distance from scapula to spinous process: Based on the key points of the scapula on both sides detected by the key point detection model, calculate the distance from the key points of the scapula to the line connecting C7 and L5 to obtain the symmetry of the distance from scapula to spinous process;

[0068] Step 4.1.7: Based on the waist concave key points detected by the key point detection model, with the waist concave points on both sides as the center, and based on the contour image calculated in 4.1.5, search upward / downward to obtain the waist concave point set, perform Hough circle fitting on the waist concave on both sides respectively, calculate the curvature of the fitting circle, and compare to obtain the waist concave symmetry index;

[0069] Step 4.1.8: Based on the waist concave key points detected by the key point detection model, combined with the contour of 4.1.5, extend to the left waist concave point and the right waist concave point respectively. The intersection with the contour is the reference point for calculating the distance from the waist concave to the arm. The distance from the reference points on both sides to the waist concave is calculated, and the symmetry index of the distance from the waist concave to the arm is obtained by comparison;

[0070] Step 4.1.9: According to the key points of the left and right iliac crests detected by the key point detection model, the angle between the connecting line and the horizontal line is calculated to obtain the inclination degree of the iliac crest;

[0071] Step 4.1.10: Based on C7 and L5, train the Yolov8 key point model to obtain 5 key points of the epidermal spines, such as Figure 3 As shown, according to the calculated 5 key points, the epidermal spinous process connection line is fitted to calculate the scoliosis angle;

[0072] Step 4.1.11: According to the coordinates of the color image of the key points T1-T12 calculated in 4.1.2, map them to the 3D point cloud, obtain their corresponding X, Y, and Z coordinates, fit a circular arc in the plane formed by the Y and Z coordinates, and calculate the kyphosis angle. When the kyphosis angle is too large and exceeds 45 degrees, there is "excessive kyphosis";

[0073] Step 4.1.12: Map the coordinates of the color image of the key points L1-L5 calculated in 4.1.2 to the 3D point cloud to obtain their corresponding X, Y, and Z coordinates, fit a circular arc in the plane formed by the Y and Z coordinates, and calculate the lordosis angle. If the lordosis angle is too large and exceeds 45 degrees, there is "excessive lordosis";

[0074] Step 4.2: Use the YOLOv8 model as the frontal coronal plane key point detection model, annotate the collected images using annotation software, and divide them into training set, validation set and test set, train the annotated training set to obtain the trained frontal coronal plane key point detection model; the key points detected by the frontal coronal plane key point detection model include: left and right greater trochanters of the femur, left and right knee joints, left and right ankles, etc.

[0075] Step 4.2.1: Leg length difference: Compare the Y coordinates of the key points of the greater trochanter of both sides to determine whether there is leg length difference;

[0076] Step 4.2.2: Knee varus: Calculate the angle based on the line connecting the greater trochanter of the femur and the knee joint and the line connecting the knee joint and the ankle to determine whether the angle is around 180°. If it is significantly less than 180°, it is knee varus. On the contrary, if it is significantly greater than 180°, it is knee valgum.

[0077] Step 4.3 Sagittal plane parameters: key point detection algorithm YOLOv8, collect images, annotate them with annotation software, and divide them into training set, validation set and test set. Train the annotated training set to obtain the trained key point detection model. The key points include: tragus, cervical 7 (C7), apex of posterior axillary line, popliteal fossa, ankle, heel, toe and other key points;

[0078] Step 4.3.1: Head forward: Based on the tragus key points and C7 detected by the key point detection model, calculate the angle between the line connecting the two key points and the horizontal line. If the angle is significantly smaller than that of a normal person (about 70°), it is considered as head forward tilt;

[0079] Step 4.3.2: Round shoulder: According to the round shoulder key points detected by the key point detection model, the Canny operator is used to detect the edge of the image. The round shoulder key points are extended to the right along the X direction, and the intersection with the edge is used as the reference point. The distance from the reference point to the round shoulder point is calculated. If the distance is too large, it is a round shoulder;

[0080] Step 4.3.4: Knee hyperextension: Calculate the angle between the line connecting the key point of the popliteal fossa and the heel and the vertical line. If the angle is greater than 5°, it is judged as knee hyperextension;

[0081] Step 4.3.5: Pointed feet: Based on the key points of the toes and heels detected by the key point detection model, measure the angle between their connecting line and the horizontal line. If the angle is too large, it is judged as pointed feet.

[0082] Furthermore, the specific method of step 1 is:

[0083] Step 1.1: The subject stands in the center of the turntable, with his back facing the depth camera and fixed in an upright position;

[0084] Step 1.2: The depth camera collects visible light and 3D images;

[0085] Step 1.3: Rotate the turntable 90°;

[0086] Step 1.4: Repeat 1.2-1.3 until 360° three-dimensional image acquisition is completed.

Claims

1. A method for integrated detection of spine, posture and sole, the method include: Step 1: Use a depth camera to collect visible light images and 3D point cloud image data of the subject in multiple positions and angles; Step 2: When the back of the person is facing the camera, start the foot scan at the same time; Step 2.1: The subject stands still, places both feet in the scanner, and adjusts the position and angle; Step 2.2: Obtain three-dimensional data of the foot and transfer the data to a computer; Step 3: Acquire coronal and sagittal dynamic back images; Step 3.1: Collect visible light and 3D images of the coronal plane and left and right lateral flexion; Step 3.2: Acquire visible light and 3D images of sagittal flexion; Step 4: Calculate relevant index parameters according to the collected images and output the calculated relevant parameters; Step 4.1: YOLOv8 is used as the back coronal plane key point detection model. The collected images are annotated with annotation software, and divided into training set, validation set and test set. The annotated training set is trained to obtain the trained back coronal plane key point detection model. The key points detected by the back coronal plane key point detection model include: left and right tragus, left and right posterior axillary line vertices, left and right scapulae, left and right lumbar concavities, left and right iliac crests, left and right popliteal fossa, left and right ankles, left and right heels, the midpoint of the back neck C7, and the midpoint of the waist L5. Step 4.1.1: Based on the key points of the head detected by the key point detection model, i.e., the left and right tragus points, the head tilt angle is measured according to the angle between the line connecting the two points and the horizontal line, and the tilt angle is output; Step 4.1.2: According to the key points C7 and L5 detected by the key point detection model, the spine is segmented and evenly divided into the midpoint of the back neck C7, the intersection of the line connecting the inner superior angles of the two shoulder blades and the longitudinal axis of the spine, and the waist. According to the length of the segment, with C7 as the origin, the coordinates of the key points are searched upward, mapped to the 3D point cloud, and the lumbar curve is fitted according to the point cloud to calculate the cervical lordosis angle; Step 4.1.3: Based on the key points of the left and right shoulder blade heights detected by the key point detection model, use the Y coordinates of the key points to perform height comparison, calculate the angle between the connecting line and the horizontal line, that is, the degree of inclination, and output the degree of inclination; Step 4.1.4: According to the key points of the left and right scapula detected by the key point detection model, map them to the 3D point cloud according to their X and Y coordinates, obtain their 3D coordinates (X, Y, Z), calculate the scapula concavity according to the difference in the Z coordinates, and output the scapula concavity; Step 4.1.5: Use the Canny operator to perform edge detection on the color image. According to the key points of the posterior axillary lines on both sides detected by the key point detection model, extend upward to make a Y-direction reference line. The intersection with the image contour is the acromion. According to the angle between the connecting line of the acromion key points and the horizontal line, calculate the inclination of the shoulder and output it; Step 4.1.6: Calculate the symmetry of the distance from scapula to spinous process: According to the key points of the scapula on both sides detected by the key point detection model, calculate the distance from the key points of the scapula to the line connecting C7 and L5, obtain the symmetry of the distance from scapula to spinous process and output it; Step 4.1.7: Based on the waist concave key points detected by the key point detection model, with the waist concave points on both sides as the center, and based on the contour image calculated in 4.1.5, search upward / downward to obtain the waist concave point set, perform Hough circle fitting on the waist concave on both sides respectively, calculate the curvature of the fitting circle, compare and obtain the waist concave symmetry index and output it; Step 4.1.8: Based on the waist concave key points detected by the key point detection model, combined with the contour of 4.1.5, extend to the left waist concave point and the right waist concave point respectively, and the intersection with the contour is the reference point for calculating the distance from the waist concave to the arm. Calculate the distance from the reference points on both sides to the waist concave, compare and obtain the symmetry index of the distance from the waist concave to the arm and output it; Step 4.1.9: According to the key points of the left and right iliac crests detected by the key point detection model, the angle between the connecting line and the horizontal line is calculated to obtain the inclination of the iliac crest and output it; Step 4.1.10: Based on C7 and L5, train the Yolov8 key point model to obtain the 5 key points of the epidermal spinous process. According to the calculated 5 key points, fit the epidermal spinous process connection line, calculate the scoliosis angle and output it; Step 4.1.11: According to the coordinates of the color map of the key points T1-T12 calculated in 4.1.2, map them to the 3D point cloud, obtain their corresponding X, Y, and Z coordinates, fit a circular arc in the plane formed by the Y and Z coordinates, calculate the kyphosis angle, and when the kyphosis angle is too large and exceeds 45 degrees, there is "excessive kyphosis", and output the detection result; Step 4.1.12: Map the coordinates of the color image of the key points L1-L5 calculated in 4.1.2 to the 3D point cloud to obtain their corresponding X, Y, and Z coordinates, fit an arc in the plane formed by the Y and Z coordinates, and calculate the lordosis angle. If the lordosis angle is too large and exceeds 45 degrees, there is "excessive lordosis", and output the detection result; Step 4.2: Use the YOLOv8 model as the frontal coronal plane key point detection model, annotate the collected images with annotation software, and divide them into training set, validation set and test set, train the annotated training set, and obtain the trained frontal coronal plane key point detection model; the key points detected by the frontal coronal plane key point detection model include: left and right greater trochanters of the femur, left and right knee joints, and left and right ankles; Step 4.2.1: Leg length difference: Compare the Y coordinates of the key points of the greater trochanter of both sides to determine whether there is leg length difference and output the detection result; Step 4.2.2: Knee valgus: Calculate the angle based on the line connecting the greater trochanter of the femur and the knee joint and the line connecting the knee joint and the ankle, and determine whether the angle is around 180°. If it is significantly less than 180°, it is knee valgus. On the contrary, if it is significantly greater than 180°, it is knee valgus, and the test result is output; Step 4.3 Sagittal plane parameters: key point detection algorithm YOLOv8, collect images, annotate them with annotation software, and divide them into training set, validation set and test set. Train the annotated training set to obtain the trained key point detection model. The key points include: tragus, cervical 7 (C7), apex of posterior axillary line, popliteal fossa, ankle, heel and toe; Step 4.3.1: Head forward: Based on the tragus key points and C7 detected by the key point detection model, calculate the angle between the line connecting the two key points and the horizontal line. If the angle is significantly smaller than that of a normal person, it is considered as head forward tilt and the detection result is output; Step 4.3.2: Round shoulder: According to the round shoulder key points detected by the key point detection model, the Canny operator is used to detect the edge of the image. The round shoulder key points extend to the right along the X direction, and the intersection with the edge is used as the reference point. The distance from the reference point to the round shoulder point is calculated. If the distance is too large, it is a round shoulder, and the detection result is output; Step 4.3.4: Knee hyperextension: Calculate the angle between the line connecting the key point of the popliteal fossa and the heel and the vertical line. If the angle is greater than 5°, it is judged as knee hyperextension and the detection result is output; Step 4.3.5: Pointed feet: Based on the key points of the toes and heels detected by the key point detection model, measure the angle between their connecting line and the horizontal line. If the angle is too large, it is judged as pointed feet and the detection result is output.

2. A method for integrated detection of spine, posture and sole as claimed in claim 1, It is characterized in that The specific method of step 1 is: Step 1.1: The subject stands in the center of the turntable, with his back facing the depth camera and fixed in an upright position; Step 1.2: The depth camera collects visible light and 3D images; Step 1.3: Rotate the turntable 90°; Step 1.4: Repeat 1.2-1.3 until 360° three-dimensional image acquisition is completed.

Images